1. Field of the Invention
The invention relates to systems and processes which use electrically generated gas plasma. The invention is particularly applicable to systems and processes which utilize a plasma in manufacturing solid state and/or semiconductor devices.
2. Discussion of the Background
Many semiconductor or solid state manufacturing processes utilize a gas plasma to perform a fabrication step. This step can be, for example, a chemical modification or an etching of a thin film, and may use chlorine gas or oxygen, among others. Often, particularly in the semiconductor industry, extremely precise control of the reaction conditions and the timing of the reaction is required. It is therefore important to accurately monitor the plasma conditions, the condition of the equipment, and the progress of the reaction.
Conventional plasma reaction systems have used optical spectrometry of the optical light emissions from the plasma in order to detect various chemical species produced by the reaction between the substrate and the plasma. Concentrations of these species can be used as an indication of the plasma conditions, or as an indication of the progress of the process being performed with the plasma (e.g., to determine the end point of the process). However, this technique is not sufficiently precise for certain processes. In particular, the optical detection method does not always provide a sufficiently reliable and timely indication of when a processing step (such as the etching of a photoresist or a metal layer) is complete.
Conventional plasma processes systems have also used timing to determine the end of a step. Several test runs are performed under the conditions to be utilized in an actual manufacturing run, in order to determine the reaction rate. During the actual manufacturing run, the plasma processing step is performed for a predetermined amount of time, and then terminated. However, slight variations in, e.g., the ambient environment, the manufacturing equipment, the plasma, and the workpiece can vary the reaction rates/times, thus making this method less than optimal.
Particularly where the processing step requires only a short amount of time, accurate determination of the end point can be crucial. For example, during etching of an extremely thin layer, a delay in the termination of the step can result in the plasma etching into the layer beneath the layer for which etching is intended. As semiconductor processes require increasingly thinner films, and as high density plasma systems allow for shorter etching and reaction times, it has become increasingly important to accurately determine the time at which a processing step is complete.
In addition, integration densities of semiconductor devices have continued to increase, demanding new levels of precision in controlling the reactive processes used for producing the devices. During a processing step, it is increasingly important to carefully control/monitor the conditions of the plasma (e.g., ion density and gas mixture), which directly impact not only rates of reaction, but also thin film material properties. System conditions, such as the cleanliness of the system, proper assembly/configuration of electrical/RF connections, RF matching, age of components, etc., can also affect the plasma conditions. Accordingly, a system and method are needed for accurately monitoring and controlling plasma and system conditions during a given process step, and for detecting the completion of the process step in order to shut down the plasma process at an appropriate time.
Known references have discussed using the harmonic content of an electrical signal to determine the condition of the plasma by exploiting the inherent non-linearity of the plasma. See Miller and Kamon (U.S. Pat. No.5,325,019 (hereinafter xe2x80x9cthe ""019 patentxe2x80x9d)), Gesche and Vey (U.S. Pat. No. 5,025,135 (hereinafter xe2x80x9cthe ""135 patentxe2x80x9d)), Patrick et at. (U.S. Pat. No. 5,474,648 (hereinafter xe2x80x9cthe ""648 patentxe2x80x9d)), Turner et al. (U.S. Pat. No. 5,576,629 (hereinafter xe2x80x9cthe ""629 patentxe2x80x9d)), and Williams and Spain (U.S. Pat. No. 5,472,561 (hereinafter xe2x80x9cthe ""561 patentxe2x80x9d)).
The ""135 patent discloses a high-pass filter of a sampled electrical signal wherein the presence of high frequency content determines the existence of a plasma. The ""019 patent uses the fundamental and harmonic frequency components of voltage and current measurements (measured at an electrode within the plasma electrical system) to select operating conditions. However, it does not teach correlating harmonic content with plasma process input parameters, such as pressure, RF input, etc. Furthermore, it does not teach control functions based upon the linear and/or non-linear combination of harmonic amplitude ratios (ratio to the amplitude of the fundamental frequency). Likewise, none of the ""019, ""629, and ""561 patents recognize that information regarding a plasma process can be obtained by modulating the RF power.
Several patents address some aspects of intelligent control of plasma processes. In particular, some patents attempt to characterize the plasma system performance, generate a database, monitor electrical components during run conditions, and compare to the database to determine the plasma conditions. For example, Kochel (U.S. Pat. No. 4,043,889) addresses this issue. It discloses a method of using a pre-determined bias voltage versus pressure characteristic to tune a process to xe2x80x98optimalxe2x80x99 conditions (in a chamber performing RF sputtering of a thin film). Moreover, Tretola (U.S. Pat. No. 4,207,137), describes controlling a plasma process.
Additionally, several patents teach monitoring the electrical properties of a plasma system and correlating their variation with plasma conditions. For example, Patrick et al. (U.S. Pat. No. 5,474,648) discloses (a) a control method to improve repeatability and uniformity of process and (b) monitoring the power, voltage, current, phase, impedance, harmonic content and direct current bias of the RF energy transferred to the plasma. Additional references describing electrical property characterization for plasma processing devices include Logan, Mazza and Davidse, xe2x80x9cElectrical characterization of radio-frequency sputtering gas discharge,xe2x80x9d J. Vac. Sci Technol., 6, p. 120 (1968); Godyak, xe2x80x9cElectrical characteristics of parrallel-plate RF discharges in Argon,xe2x80x9d IEEE Transactions on Plasma Sci., 19(4), p. 660(1991); and Sobolewski, xe2x80x9cElectrical characterization of radio-frequency discharges in the Gaseous . . . xe2x80x9d, J. Vac. Sci. Technol., 10(6) (1992). For real-time control of etching processes using multi-variate statistical analysis, see Fox and Kappuswamy (U.S. Pat. No. 5,479,340).
Some patents also discuss monitoring the optical properties of a plasma. Using an optical emission spectrometer, information about the species present within the plasma (and their approximate concentration) can be ascertained from monitoring the emission spectrum of the light present. In fact, several spectrometers (or those which comprise a rotary grating) may monitor the presence of several species and, hence, provide a plurality of inputs to a plasma process control system. See Cheng (U.S. Pat. No. 5,160,402) and Khoury, Real-time etch plasma monitor system, IBM Technical Disclosure Bull., 25(11A) (1983).
Turner (U.S. Pat. No. 4,166,783) proposes a computer control system for use with deposition rate regulation in a sputtering chamber. The system records the use of the sputtering device and compiles a history of its performance. During future use of the device, the past performance, age, etc., are incorporated into adjustments made during a run condition.
Automatic impedance matching systems are also known which employ a (quasi-)intelligent controller to monitor an electrical property of the plasma chamber. In fact, some systems attempt to obtain some correlation between settings for variable reactances (i.e., capacitors and inductors) and plasma conditions such as the load impedance or plasma chamber input parameters (i.e., RF input power, chamber pressure, etc.). If correlation is obtained, then coarse tuning of the impedance matching also can be obtained. For example, U.S. Pat. No. 5,195,045 to Keane and Hauer presents a method of using pre-determined set points for two impedance varying devices in order to solve tuning problems during run conditions. Additionally, U.S. Pat. No. 5,543,689, to Ohta and Sekizawa, proposes storing match circuit settings from prior use. U.S. Pat. No. 5,621,331, to Smith et al., presents a method for rapidly adjusting the impedance of a variable impedance device to match (a) the impedance of a source to (b) the impedance of a load in a plasma processing device. The device includes a plurality of electrical sensors, a photosensitive detector, a data processor, and a memory. The ""331 patent correlates (1) variable reactance settings and (2) measurements of chemical species present within the plasma using an optical emission spectrometer and electrical measurements taken on plasma coupling elements. In this manner, a set of chemistry conditions may be selected and tuned by monitoring the variable reactance settings.
Neural networks have been used for both predication and control in many areas. A use of neural networks in semiconductor processing to predict the endpoint of an etch process is discussed by Maynard et al. in xe2x80x9cPlasma etching endpointing by monitoring RF power systems with an artificial neural network,xe2x80x9d Electrochem. Soc. Proc., 95-4, p189-207, 1995, and xe2x80x9cPlasma etching endpointing by monitoring radio-frequency power systems with an artificial neural network,xe2x80x9d J. Electrochem. Soc., 143(6).
In view of the foregoing, it is an object of the invention to provide a system and method which can quickly and accurately monitor and control a process that uses a gas plasma, e.g., in a processing system which generates a plasma using radio-frequency (RF) power.
According to one aspect of the invention, the plasma conditions and system conditions are sensed by varying or modulating one or more of the amplitude, phase, and frequency of RF power on an element in the system, and by observing characteristics of the resulting response signals of the same element, and/or one or more other elements in the system. The xe2x80x9celementsxe2x80x9d are electrical components of the system, more particularly plasma coupling elements such as electrodes, a bias shield, an inductive coil or an electrostatic chuck. The electrical components can also include a probe or another sensor utilized to obtain signals in response to modulation of the power.
In one exemplary embodiment, each of the modulated signals is observed at a node of a power delivery circuit, which can include, for example, an electrical matching network. The measured response information is then compared with stored data obtained for known conditions, and the system conditions are determined as corresponding to the conditions under which the stored data (which most closely resembles the measured data) was obtained.
In a presently preferred embodiment, in order to meaningfully utilize information obtained by modulating the power to one or more of the system components, the processing system is characterized with a series of test runs. The term xe2x80x9ctest runxe2x80x9d is used herein to mean a test of a processing system performed for the purpose of characterizing the system. The test run can be performed under conditions similar to those under which a production run would be performed. In addition, the test run can correspond to conditions to be utilized in performing a diagnostic test of equipment to be utilized in a processing run, so that information obtained in the test run can be utilized to determine the condition of equipment, for example, before a process step commences. A test run is performed by operating the system under a given/known set of plasma conditions and system conditions, applying modulated RF power to one or more of the electrical components (more particularly the plasma coupling elements), and measuring the resulting modulation response signals. A series of test runs are performed by systematically varying the plasma conditions and/or system conditions so that response signals are obtained and correlated with the various conditions, thereby providing response signal data which can be stored as characteristic of the known conditions. For example, the power, pressure, gas mixture, or material being processed can be varied, and the effects of these variations on the modulation response signals can be determined empirically. Additional conditions varied during a test/characterizing run can include cleanliness (time or cycles since cleaning/maintenance was performed), age of components, ambient temperature and/or humidity, etc. It is to be understood that the number of condition variables for which data is obtained can vary depending upon, e.g., the extent to which the profiles will be utilized throughout a processing run, level of sophistication/control desired, and budgetary constraints. For example, a relatively simple system might utilize modulation response information only to detect the end point of a process, or to indicate the cleanliness of a system. A more complex system can consider additional conditions/variables throughout a process.
The response information obtained during a test run forms a database, which includes a multi-dimensional array of data points, each data point containing information regarding a characteristic, such as the amplitude or phase, of one modulated signal component. For a given set of plasma and system conditions, the data points can be grouped together to form a modulation profile. The term xe2x80x9cmodulation profilexe2x80x9d is used herein to mean a set of data associated with a set of process/system conditions, which represents the characteristics of the signals caused by modulating the RF power to one or more plasma coupling elements. Each set of process conditions stored in the database is thereby associated with (i.e., linked to) a modulation profile.
The data obtained during the test run is then utilized during an actual production run, in which the process conditions can be determined/inferred by modulating the RF power on various plasma coupling elements, observing the effects of the modulation on signals from the same, or other, plasma coupling elements, and constructing an observed modulation profile from these measurements. To determine the current process conditions, the system uses either (1) the stored modulation profile (i.e., the profile from the database obtained during test runs) that most closely matches the observed modulation profile, or (2) a neural network which predicts the current process conditions from the observed modulation profile. When using the first method, the database links this stored profile to a set of stored process conditions. The current process conditions are thus determined/inferred, based on these stored process conditions. For example, in a plasma etching system having a plasma coupling element, the RF power being supplied to this element can be amplitude modulated at a certain modulation frequency and modulation amplitude. The resulting modulation response signal is observed by measuring the voltage at, e.g., a node of the power delivery circuit. The amplitude of the modulation response signal can have one value during the etching of a film and a second value after the etching has been completed, based upon empirical data obtained in a test run. The plasma state (and therefore, the process state) during etching is associated with a first modulation profile, and the plasma state (and the process state) after etching is associated with a second modulation profile. Each of the profiles in this example would contain a single data point, corresponding to the amplitude of the modulation response signal. During a processing run, the modulation response signal can then be utilized to determine the end point of an etching process, i.e., when the second modulation profile is detected. When using the second method, the neural network determines the end point of an etching process. An observed modulation profile is applied as an input to the neural network to determine when etching is complete without requiring an exact match of a modulation profile.
According to the present invention, it has been recognized that the characteristics of the aforementioned modulation response signals can be sensitive to changes in plasma and system conditions, and the progress of a reaction. Consequently, a more accurate or complete xe2x80x9cpicturexe2x80x9d of the process conditions can be obtained as compared with conventional systems. In addition, since process/system conditions can be extremely well-characterized, deviations from the desired process conditions can be accommodated or cancelled out by modifying the characteristics of the RF power. Further, when using a neural network, small errors are naturally compensated for by the neural network. Alternatively, when it is determined that conditions are unsuitable, corrective action can be taken, while preventing or minimizing damage to the equipment or substrates being processed. Therefore, improved consistency/repeatability of the process and resulting product is achieved. According to another aspect of the invention, a wide variety of process conditions and/or observed modulation effects/responses can be displayed on a monitor screen, so as to rapidly give an operator useful information regarding the process or status of the process.
The invention offers several advantages over conventional systems. In particular, by modulating the RF power and observing the modulation response signals in the electrical circuits of the system, the current characteristics of a plasma can be determined to a greater degree than optical detection or timing methods used in conventional systems. In addition, the invention allows plasma processing systems to be thoroughly tested and characterized under well-controlled conditions before delivery or use in manufacturing substrates, so as to provide quality assurance of the systems, as well as insuring the accuracy of the database used to analyze the modulated signals. These features allow for more precise control of the manufacturing processes, thereby improving the quality assurance of the wafers (or other substrates) being produced. Moreover, improving the quality assurance results in fewer defective wafers, so that the manufacturing yield is increased and manufacturing costs are decreased. Furthermore, the information obtained by the modulation technique can be displayed in an easily-understandable format, on a monitor screen, in order to allow an operator to accurately monitor the system or a process being performed and to detect problems. This results in improved reliability and reduced risk of operator error.